Research And Implementation Of Federated Learning For Secure Data Sahring Of IoV

With the rapid development of IoV,IoV information services are constantly enriched and improved.The realization of IoV information service requires the use of driving manipulation data,environmental data and user data collected by different vehicles.These vehicle data contain a large amount of user personal information.Therefore,centralized uploading and processing vehicle data poses a greater risk of privacy leakage.Federated learning converts the model training process of centralized machine learning into four key steps:local model training,local model upload,global model aggregation and global model delivery.Federated learning can use multiple data sets to jointly train service models without the original data leaving the device,which achieve the goal of data secure sharing.The prospect of federated learning in the development of data sharing in IoV is promising,but it also faces many challenges.Two key issues need to be addressed.The first one is the reduced efficiency of federated learning caused by vehicles’ characteristics.Compared with other static scenarios,the impact of strong mobility and resource heterogeneity of vehicles needs to be considered.The second one is the data privacy issue in the sharing process.While local model training guarantees that the original dataset is stored locally,the interacting intermediate model parameters still runs the risk of leaking the original data.In order to better protect user privacy,privacy protection measures need to be added in the interaction process,but the loss of model accuracy and training efficiency may not meet the task requirements.In response to the above challenges,this paper conducts in-depth research and discussion on how to efficiently realize the data sharing of IoV while protecting data privacy,and completes the following work:1.In view of the decline of federated learning efficiency caused by vehicle characteristics,this paper refines the evaluation dimension of vehicle heterogeneous resources and proposes a resource-heterogeneous participating vehicle selection algorithm.Firstly,the node stability of the vehicle is screened,and then the vehicle selection problem is transformed into the problem of maximizing the number of participants within the time threshold.A solution based on a greedy algorithm is proposed to select suitable vehicle nodes to participate in model training.Experiments show that,compared with the existing random selection method and FedCS algorithm,the proposed algorithm can effectively improve the model training efficiency of federated learning.2.Aiming at the problem of privacy leakage in vehicle data sharing,this paper proposes a federated learning method for secure data sharing of IoV.This method enhances the privacy protection capability of federated learning by means of homomorphic encryption and differential privacy technology,thereby effectively protecting the intermediate model parameters.In order to solve the problem of the loss of model accuracy and the decline of training efficiency caused by additional privacy protection measures,which cannot meet the task requirements,this paper adaptively selects differential privacy and homomorphic encryption according to the task and vehicle configuration.Experiments show that the method proposed in this paper can achieve safe and efficient shared model training within an acceptable resource consumption range according to the actual situation.